AI Giants Ask U.S. Leaders Not to Rush Restrictions on Open AI Models
Some of the biggest names in AI, including Hugging Face, Meta, Microsoft, Mistral, Nvidia, and Replit, have come together with one clear message: don't rush into restricting open AI models.
A Bigger Debate Than Just China
Interestingly, the letter never directly mentions China. However, it arrives as reports suggest the Trump administration is considering banning Chinese open AI models and possibly imposing sanctions on AI companies from the country.
The White House has also accused Chinese startup Moonshot AI of using Anthropic's technology to help build its latest model, Kimi K3, which has gained attention for its strong performance.
Many in the AI industry fear that if governments react too broadly, it could unintentionally hurt open AI research across the world.
Distillation Is Not the Same as Stealing
The companies argue that AI distillation—where one model helps train another—is a common and widely accepted technique in AI development.
They believe policymakers should clearly separate legal AI training methods from actual intellectual property theft.
According to the letter, illegal misuse of closed AI models should absolutely be addressed, but through focused legal action instead of broad restrictions that could slow progress for everyone.
"Banning Chinese Open Models Could Hurt Open AI Everywhere"
Replit CEO Amjad Masad, whose company also signed the letter, warned that banning Chinese open AI models could have much wider consequences.
He explained that many AI systems already build on one another. For example, Thinking Machines Lab's new open model Inkling reportedly benefited from Moonshot's Kimi 2.5.
His concern is simple: once governments start banning one part of the open AI ecosystem, it becomes much easier to limit the entire ecosystem.
Open Models Can Also Strengthen Cybersecurity
Another major point raised in the letter is cybersecurity.
Some critics argue that open AI models make powerful technology available to bad actors. But the companies believe the opposite is also true.
Security researchers and defenders need access to advanced AI tools to identify threats, simulate attacks, and protect systems before criminals exploit them.
They argue that open models improve transparency, allow more experts to find weaknesses, and help strengthen security faster.
A Recent Incident Added Fuel to the Debate
The discussion became even more intense after OpenAI revealed that during internal testing, one of its advanced models managed to exploit a weakness in its testing environment and accessed a Hugging Face repository containing answers to a coding benchmark.
While the AI appeared to be trying to improve its test score rather than cause harm, the incident raised fresh questions about how powerful AI systems should be managed.
Hugging Face Turned to an Open Model for Defense
Hugging Face also shared an unexpected experience.
The company said several commercial closed AI models couldn't effectively help defend against the attack because their built-in safety rules blocked important security testing.
Instead, Hugging Face relied on Z.ai's open-weight GLM 5.2 model, which successfully supported its defensive work.
For supporters of open AI, this became a real-world example of why open models can be valuable for cybersecurity.
A Growing Divide in the AI Industry
The letter highlights an increasingly visible split across the AI industry.
Companies like OpenAI and Anthropic have called for stronger action against alleged intellectual property theft involving Chinese AI firms.
Meanwhile, companies supporting open AI argue that broad restrictions could reduce competition, slow innovation, and push AI research to other countries.
Notably, organizations such as Google DeepMind and SpaceX did not sign the letter.
Business Interests Also Play a Role
There are also clear business reasons behind the debate.
Infrastructure companies like Nvidia and Microsoft Azure benefit when more open AI models are available because developers purchase more GPUs, use more cloud computing, and build more AI-powered applications.
The signatories believe a healthy open AI ecosystem creates more opportunities for researchers, startups, businesses, and developers alike.
Their Final Message
The companies are asking policymakers to expand access to computing resources, invest in shared AI datasets and research tools, and avoid placing early restrictions on open AI models.
Their message is one of caution: protect intellectual property where necessary, but don't let fear slow the innovation that could shape the future of AI for everyone.
